Architecting dependable systems using virtualization

We propose new methods of leveraging virtualization for addressing system dependability issues. Using combinatorial modeling, we analyze multiple design choices when a single physical server is used to host multiple virtual servers. Our results show that unless certain conditions (e.g., regarding the reliability of the hypervisor and the number of VMs) are met, virtualization could decrease the reliability of a single physical node. In light of the prevailing ad-hoc approach to virtualization and the general inclination to move services out of the operating system into the virtualization layer, our results point out the need for a more cautious and rigorous approach.